search

brian-j-smith / Mamba.jl

Markov chain Monte Carlo (MCMC) for Bayesian analysis in julia
Other
254 stars 52 forks source link

Un-metropolize BMG #69

Closed bdeonovic closed 8 years ago

bdeonovic commented 8 years ago

After reading several papers including a closer inspection on Schafer's paper I realized nobody ever intended for this sampler to use a Metropolis-Hastings step; its just a Gibbs sampler. I've updated the sampler, documentation, and name correspondingly.

brian-j-smith commented 8 years ago

I think it's fine to leave the BMG sampler in as an implementation of Algorithm 1 in the Schafer paper. I plan to update the BMG code to match that algorithm, including an option for user-specified number of parameters to update at each call (similar to BMMG). The proposed BGS can be viewed as a special case of the existing DGS. So, rather than creating a new binary Gibbs sampler, maybe consider extending the DGS one as needed.